A multi-dimensional fusion strategy similarity measure method for patent application technology disclosure document

Area of Science:

• Intellectual Property Management
• Computer Science
• Information Retrieval

Background:

• Automated evaluation of patent application technology disclosure documents is crucial for assessing novelty and uniqueness, but current methods face challenges.
• Existing text similarity algorithms struggle with limited cross-library data and insufficient focus on core content, hindering practical application.
• Human evaluation is time-consuming and subjective, necessitating more objective automated solutions.

Purpose of the Study:

• To propose a novel multi-dimensional fusion strategy for similarity evaluation of patent application technology disclosure documents.
• To enhance the accuracy and efficiency of automated patent similarity assessment.
• To provide a new method for reliable patent novelty and uniqueness judgment.

Main Methods:

• Developed text preprocessing strategies including word segmentation reconstruction and weighted fusion of word frequency and part-of-speech scores.
• Introduced a similarity calculation method utilizing dot matrix and image mapping spaces for diversified evaluation.
• Evaluated the proposed algorithm on published text similarity datasets and a specific dataset of patent application technology disclosure documents.

Main Results:

• The multi-dimensional fusion strategy improved discrimination accuracy by approximately 10% compared to traditional vector semantic models on benchmark datasets.
• The method achieved comparable discriminative ability to lightweight deep learning models without requiring training.
• On patent disclosure document datasets, the proposed method outperformed traditional models by 20% and deep learning models by 1-8%, with balanced precision and recall.

Conclusions:

• The proposed multi-dimensional fusion strategy offers a significant advancement in evaluating patent application technology disclosure document similarity.
• The dot matrix and image space mapping methods provide effective and reliable similarity calculations.
• This research presents a valuable new approach for automated patent similarity evaluation, improving upon existing techniques.